The development of new materials, incorporation of new functionalities in materials, and even the description of well-studied materials strongly depends on the capability to understand and predict complex structure-properties relationships. A significant challenge in this field remains the near infinite size of the chemical/materials/process spaces, which makes an exhaustive characterization of materials even with a rather moderate number of constituents infeasible.

To go beyond this state-of-the-art, computational approaches increasingly complement experiment, but to do so effectively, these methods must go beyond explanation towards prediction (1,2). This has been achieved in some areas, most notably for ground-state electronic properties (3), but there is still a long way to go to model materials in the context of their materials system and process chain (4).

Simulation and ML can produce digital twins of specific that can be connected in a modular library to form powerful computational workflows (5). Such workflows need to combine different computational methods on different scales such as DFT calculations on the quantum level, MD simulations on the atomistic level and FEM on the continuum scale and increasingly ML/AI methods (6-10). These efforts at model development must be complemented by efficient numerical implementations which can be combined to complex computational workflows.

Work in this direction (see references) has led to numerous breakthroughs, but also led to the realization that simulations/AI often cannot go all the way. Predicted materials are often difficult to synthesize, lack stability or interact with the environment in unforeseen ways. The implementation of materials acceleration platforms (MAP) (11-14) is one of the most recent developments to overcome this predicament. In these MAP the experiment is organized into a sample-centric fashion, which permits an increasingly automated synthesis and characterization of the material with tests of its functionality in the context of applications. While these robotic setups can increase the experimental throughput, fail to address the size of the materials search space. Experiment must therefore be complemented in a closed-loop fashion with predictive computational tools, either simulation or AI driven, that narrow the search space to the most promising regions.

Invited talks:

Christoph Brabec, FAU Erlangen, D, A hybrid data-based – knowledge-based workflow to discover tailored energy materials

Gábor Csányi, University of Cambridge, UK, Foundation models for atomistic materials chemistry

Gianaurelio Cuniberti, TU Dresden, D, Machine Learning for Molecular Sensing

Volker Deringer, University of Oxford, UK

Glenn H. Fredrickson, University of California, USA, Predictive Multiscale Modeling of Polymer Formulations

Maciej Haranczyk, IMDEA Materials Institute, E

Tilmann Hickel, MPI for Sustainable Materials, D

Kim Jelfs, Imperial College London, UK, Remembering the lab in computational materials discovery

Mark Kozdras, University of Toronto, CDN

Boris Kozinsky, Harvard, USA

Kurt Kremer, MPI for Polymer Research, D

Takashi Miyake, National Institute of Advanced Industrial Science and Technology (AIST), J, Finite-temperature properties of magnetic materials

Karsten Reuter, Fritz Haber Institute, MPI, D, Accelerated Materials Discovery for Energy Conversion and Storage

Stephan Roche, ICREA, E, Exploring properties and applications of amorphous 2D materials in nanoelectronics using Artificial Intelligence

Egbert Zojer, TU Graz, A, Understanding thermal properties of complex materials using machine-learned force fields with ab initio accuracy